30,000 DICOM headers needed to be mined from PACS. How?

arongycan · Jan 22nd 2019

Hello folks,

I would need to do a big data mining.

To access PACS system based on 30,000 accession numbers, and download DICOM headers for these patients.

Does anyone have any idea how to do this without downloading the whole DICOM file (which is obviously not feasible based on the numbers of patients)?

Thank you for any idea!!!

marcelvanherk · Jan 22nd 2019

Hm,

obviously you can query the PACS, but that will not give you teh full header. A modern PACS system may have web service to get the header data. Another option is to retrieve one slice per series. On our PACS that is still slow though, because data is moved to faster storage if you get even a bit of it. If the data is in Conquest I could help you out further.

Marcel

arongycan · Jan 23rd 2019

ok, thanks for the reply!

Anyhow, how to solve that these examinations are over period of ten years, and some are in PACS, but some are in long term archive. I cannot stop hospital servers working.

Is it possible somehow to access examinations in archives and read just their full headers?

Pushing one slice per serie would be acceptable option, because then I can use dcmdump utility and strip DICOM headers.

However, is it doable by some script to pull one slice per serie of one examination automatically for 30,000 examinations?

Thx!

marcelvanherk · Jan 23rd 2019

Yes,

run conquest DICOM server as client, and write a lua script where you query for all studies then series, and then per series query for all images, and then move a selected image. I have a sample at work I can post.

Marcel

arongycan · Jan 23rd 2019

Marcel, that sounds very promising!

We have an appointment with our IT PACS team about this issue.

However, if you are willing to share your sample with me, it will be great help for me for this discussion with IT PACS people. I will be happy to include you on the papers which we want to write out of this database.

Thanks,

Martin

marcelvanherk · Jan 23rd 2019

No Worries,

left it at work - just rewrote it:

Code

PACS = 'CONQUESTSRV1' -- for testing

-- query all studies (modify as needed)
q = DicomObject:new()
q.QueryRetrieveLevel = 'STUDY'
q.PatientID = ''
q.StudyInstanceUID = ''
r = dicomquery(PACS, 'STUDY', q)

for i=0, #r-1 do
  -- query series per study (modify as needed)
  q2 = DicomObject:new()
  q2.QueryRetrieveLevel = 'SERIES'
  q2.StudyInstanceUID = r[i].StudyInstanceUID
  q2.SeriesInstanceUID = ''
  r2=dicomquery(PACS, 'SERIES', q2)
  
  for j=0, #r2-1 do
    -- query objects per series
    q3 = DicomObject:new()
    q3.QueryRetrieveLevel = 'IMAGE'
    q3.StudyInstanceUID = r[i].StudyInstanceUID
    q3.SeriesInstanceUID = r2[j].SeriesInstanceUID
    q3.SOPInstanceUID = ''
    r3=dicomquery(PACS, 'IMAGE', q3)
    
    -- get one object (the first one returned from the query)
    q4 = DicomObject:new()
    q4.QueryRetrieveLevel = 'IMAGE'
    q4.StudyInstanceUID = r[i].StudyInstanceUID
    q4.SeriesInstanceUID = r2[j].SeriesInstanceUID
    q4.SOPInstanceUID = r3[0].SOPInstanceUID
    dicommove(PACS, 'CONQUESTSRV1', q4)

    -- read it from conquest and dump the header
    im = DicomObject:new()
    im:Read(r[i].PatientID .. ':' .. r3[0].SOPInstanceUID)
    im:Dump(q4.SOPInstanceUID..'.txt')
    print('loaded image', q4.SOPInstanceUID, 'of total images', #r3)
  end
end

Display More

This loads the header of one image per series. Easiest run it from windows.

Marcel

30,000 DICOM headers needed to be mined from PACS. How?

Participate now!

Share

Tags